Pattern Recognition Techniques in Image-Based Material Classification of Ancient Manuscripts

Classifying ancient manuscripts based on their writing surfaces often becomes essential for palaeographic research, including writer identification, manuscript localization, date estimation, and, occasionally, forgery detection. Researchers continually perform corroborative tests to classify manuscripts based on physical materials. However, these tests, often performed on-site, require actual access to the manuscript objects. These procedures involve specific expertise in manuscript handling, a considerable amount of time, and cost. Additionally, any physical inspection can accidentally damage ancient manuscripts that already suffer degradation due to aging, natural corrosion, and damaging chemical treatments. Developing a technique to classify such documents using noninvasive techniques with only digital images can be extremely valuable and efficient. This study uses images from a famous historical collection, the Dead Sea Scrolls, to propose a novel method to classify the materials of the manuscripts. The proposed classifier uses the two-dimensional Fourier transform to identify patterns within the manuscript surfaces. Combining a binary classification system employing the transform with a majority voting process is adequate for this classification task. This initial study shows a classification percentage of up to 97% for a confined amount of manuscripts produced from either parchment or papyrus material. In the extended work, this study proposes a hierarchical k-means clustering method to group image fragments that are highly likely to originate from a similar source using color and texture features calculated on the image patches, achieving 77% and 68% for color and texture clustering with 100% accuracy on primary material classification. Finally, this study explores a convolutional neural network model in a self-supervised Siamese setup with a large number of images that obtains an accuracy of 85% on the pretext task and an accuracy of 66% on the goal task to classify the materials of the Dead Sea Scrolls images.